OpenAI dots alternative you can run in the UK and EU

OpenAI launched dots on 29 September 2026. They're always-on agents running on GPT-6 Astra, each with its own cloud computer and browser, that keep working through your connected apps after you close ChatGPT. Personal Pro accounts in the UK, Switzerland and the European Economic Area can't get one yet. If you're in that group, or you'd rather keep the agent on a machine you control, Hermes Agent on a VPS covers most of the same ground.

I haven't used a dot. Everything about dots below comes from OpenAI's launch post and its help article, read the day after launch, and OpenAI hasn't published the usage terms yet, so expect the pricing part to move.

What OpenAI dots do

A dot, as the OpenAI announcement describes it, is an agent you hand ongoing work to. You give it goals and access to apps. It carries on in the background and messages you with progress or with a decision it needs from you. OpenAI's own summary fits in one line:

"Dots can do nearly anything using their own cloud computer, their own browser, and the apps you've connected."

When you aren't talking to it, a dot does what OpenAI calls proactive research. During that time its tools are restricted to read-only, so it can't send messages, change app content or control your browser. Actions that could affect your accounts go through auto-review, a check against your instructions and Custom Rules, and a few tasks (changing a password is OpenAI's example) always stay with you.

You reach a dot in the ChatGPT apps or on the web, by message or by call, and also in Slack and Teams. Texting is announced as coming soon. OpenAI counts more than 4,000 apps a dot can connect to and describes specialist dots for company roles. Teams of dots working together are described as something for later; today each account starts with one.

Who can use OpenAI dots in the UK and EU

OpenAI's getting started with your dot article is more precise than the launch coverage was. The exclusion turns out to be narrower:

"Dots are rolling out in ChatGPT on web, mobile, and desktop starting today to Pro users in markets excluding the European Economic Area, Switzerland, and the UK. Dots are also available to Business Premium users across all supported ChatGPT regions, and Enterprise users can try the beta when their workspace admin enables it."

So a personal Pro subscriber in London or Warsaw doesn't get a dot, while a company on Business Premium in the same cities does. OpenAI hasn't given a date for personal accounts in those regions.

The first dot is included with the plan. Pro starts at $100 a month on Pro 100, with Pro 200 and Pro 500 above it. Plus and the cheaper plans don't get dots at all. For the first month dots usage doesn't count against plan allowances; after that OpenAI says it will share usage terms for each plan. Extra dots, and faster or higher-capacity ones, are announced for later with no price attached.

Hermes Agent as a self-hosted dots alternative

Hermes Agent is an open-source agent from Nous Research under the MIT licence. It runs as a long-lived process on a server and keeps memory across sessions, with recurring jobs handled by a built-in cron scheduler. It reaches other tools through MCP servers configured in YAML or through skills, and bigger tasks can be split across subagents working in parallel. You talk to it through gateways for Telegram, WhatsApp, Slack, Discord, Signal and email.

The model is your choice. Hermes works with API keys from the big model providers or OpenRouter, or with a local model served from your own hardware. The project's GitHub repository lists the supported providers and gateways, and our guide to installing Hermes Agent on Ubuntu covers the install from a fresh server. After that, connect a chat app; the Telegram gateway is the quickest one to get talking. WhatsApp and email take longer, since each needs its own bridge or mailbox credentials.

I'd copy one idea from OpenAI straight into any Hermes install: keep everything the agent does unprompted on read-only tools. A morning cron job that reads your inbox and drafts a summary can't embarrass you. One that can also send mail can, and it will do it at 6am when nobody is watching.

Where your data goes with each option

A dot works on OpenAI's cloud computer. The apps you connect hand their data to it there. With Hermes, the agent's memory files and logs sit on your server's disk in the location you chose.

The model call is the exception. If Hermes talks to a hosted provider such as Anthropic or OpenAI, every prompt still leaves your server for that provider, under that provider's terms. Self-hosting moves the memory and the files onto your disk; inference stays with whoever serves the model.

Only a local model keeps the whole loop on one machine.

Running Hermes with a local model on a GPU

Hermes can point at a local Ollama server instead of a hosted API. Before the first run, raise Ollama's context window (the Hermes and Ollama context window fix walks through it), because an agent loop fills the default quickly. Model size is then limited by VRAM, which our Ollama hardware requirements guide maps model by model. A 16 GB card and a 96 GB card give you very different agents.

A GPU server costs more per month than a Pro 100 plan, so for one person running a handful of daily tasks the local route buys control and costs more. It starts to pay off when several people or several agents share the GPU, or when the data can't go to a model provider at all.

LumaDock's Hermes template installs the agent on a VPS in one click, and the same agent runs on a GPU VPS with RTX PRO Blackwell cards when you want the model on the box too.

What you take on when you self-host an agent

A dot's upkeep is OpenAI's problem. A Hermes server's upkeep is yours, and it looks like this:

• Updates: Hermes changes often, and a version bump can change config keys or gateway behaviour, so read the release notes before you upgrade.
• Model spend: the guide to cutting Hermes token costs is worth reading before the first bill arrives. Background jobs call the model even when there's nothing new to act on.
• Guardrails: there's no auto-review layer built by a model lab, so which tools the agent gets and what needs your approval is your configuration.
• Backups: the memory files are the agent. Lose the disk without a backup and you start from a stranger.

Some things a dot does that Hermes won't match without work: you can call your dot like a person, and the 4,000-app catalogue comes with the sign-in plumbing already done. On Hermes, each new app is an MCP server or a skill you install and authorise yourself.

OpenAI dots vs Hermes Agent

If you're a Pro subscriber in the US or another eligible market and you want nothing to maintain, use the dot. It's included in what you already pay and someone else handles the server.

Running a personal account from the UK or the EEA? Today the choice is a Hermes server or waiting for a date OpenAI hasn't announced. Teams on Business Premium in Europe can use dots now, and Hermes still makes sense for them when the model or the data needs to live somewhere OpenAI's cloud doesn't reach.

The number to watch is OpenAI's usage terms, due once the free first month ends around late October. Until then nobody outside OpenAI knows what heavy use of a dot costs, while a server's price is on the order page the day you buy it.